Existing pedestrian re-identification methods based on deep learning have achieved good results under constrained conditions. However, there exist some challenges including large human pose variations, viewpoint changes, severe occlusions and imprecise detection of persons. So we present a Hard/Soft hybrid Attention Network (HSAN) that combines pose information and attention mechanism to deal with the challenges. Our model includes two main parts: Pose-guided Hard Attention (PHA) and Regional Soft Attention (RSA). PHA uses the keypoints generated by pose estimation to enhance the foreground information, and RSA is learned to eliminate the background clutter. We extract reliable features and locate discriminative regions by using these two modules to handle occlusions, pose changes and background noises. We conduct a lot of experiments on public datasets including DukeMTMC-ReID, Market-1501, and CUHK03, and the results show that our method achieves stateof-the-art performance.